anv: Move INTERFACE_DESCRIPTOR_DATA setup to the pipeline

There are a few dynamic bits, namely binding table and sampler addresses,
but most of it is static and really belongs in the pipeline.  It certainly
doesn't belong in flush_compute_descriptor_set.  We'll use the same state
merging trick we use for gen7 DEPTH_STENCIL.

Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
This commit is contained in:
Jason Ekstrand
2016-11-12 12:13:17 -08:00
parent 8db6f2e6eb
commit d33e2ad67c
3 changed files with 38 additions and 27 deletions

View File

@@ -1414,6 +1414,8 @@ struct anv_pipeline {
struct { struct {
uint32_t wm_depth_stencil[4]; uint32_t wm_depth_stencil[4];
} gen9; } gen9;
uint32_t interface_descriptor_data[8];
}; };
static inline bool static inline bool

View File

@@ -1338,7 +1338,6 @@ void genX(CmdDrawIndexedIndirect)(
static VkResult static VkResult
flush_compute_descriptor_set(struct anv_cmd_buffer *cmd_buffer) flush_compute_descriptor_set(struct anv_cmd_buffer *cmd_buffer)
{ {
struct anv_device *device = cmd_buffer->device;
struct anv_pipeline *pipeline = cmd_buffer->state.compute_pipeline; struct anv_pipeline *pipeline = cmd_buffer->state.compute_pipeline;
struct anv_state surfaces = { 0, }, samplers = { 0, }; struct anv_state surfaces = { 0, }, samplers = { 0, };
VkResult result; VkResult result;
@@ -1352,9 +1351,6 @@ flush_compute_descriptor_set(struct anv_cmd_buffer *cmd_buffer)
struct anv_state push_state = anv_cmd_buffer_cs_push_constants(cmd_buffer); struct anv_state push_state = anv_cmd_buffer_cs_push_constants(cmd_buffer);
const struct brw_cs_prog_data *cs_prog_data = get_cs_prog_data(pipeline);
const struct brw_stage_prog_data *prog_data = &cs_prog_data->base;
if (push_state.alloc_size) { if (push_state.alloc_size) {
anv_batch_emit(&cmd_buffer->batch, GENX(MEDIA_CURBE_LOAD), curbe) { anv_batch_emit(&cmd_buffer->batch, GENX(MEDIA_CURBE_LOAD), curbe) {
curbe.CURBETotalDataLength = push_state.alloc_size; curbe.CURBETotalDataLength = push_state.alloc_size;
@@ -1362,31 +1358,18 @@ flush_compute_descriptor_set(struct anv_cmd_buffer *cmd_buffer)
} }
} }
const uint32_t slm_size = encode_slm_size(GEN_GEN, prog_data->total_shared); uint32_t iface_desc_data_dw[GENX(INTERFACE_DESCRIPTOR_DATA_length)];
struct GENX(INTERFACE_DESCRIPTOR_DATA) desc = {
.BindingTablePointer = surfaces.offset,
.SamplerStatePointer = samplers.offset,
};
GENX(INTERFACE_DESCRIPTOR_DATA_pack)(NULL, iface_desc_data_dw, &desc);
const struct anv_shader_bin *cs_bin =
pipeline->shaders[MESA_SHADER_COMPUTE];
struct anv_state state = struct anv_state state =
anv_state_pool_emit(&device->dynamic_state_pool, anv_cmd_buffer_merge_dynamic(cmd_buffer, iface_desc_data_dw,
GENX(INTERFACE_DESCRIPTOR_DATA), 64, pipeline->interface_descriptor_data,
.KernelStartPointer = cs_bin->kernel.offset, GENX(INTERFACE_DESCRIPTOR_DATA_length),
.BindingTablePointer = surfaces.offset, 64);
.BindingTableEntryCount = 0,
.SamplerStatePointer = samplers.offset,
.SamplerCount = 0,
#if !GEN_IS_HASWELL
.ConstantURBEntryReadOffset = 0,
#endif
.ConstantURBEntryReadLength =
cs_prog_data->push.per_thread.regs,
#if GEN_GEN >= 8 || GEN_IS_HASWELL
.CrossThreadConstantDataReadLength =
cs_prog_data->push.cross_thread.regs,
#endif
.BarrierEnable = cs_prog_data->uses_barrier,
.SharedLocalMemorySize = slm_size,
.NumberofThreadsinGPGPUThreadGroup =
cs_prog_data->threads);
uint32_t size = GENX(INTERFACE_DESCRIPTOR_DATA_length) * sizeof(uint32_t); uint32_t size = GENX(INTERFACE_DESCRIPTOR_DATA_length) * sizeof(uint32_t);
anv_batch_emit(&cmd_buffer->batch, anv_batch_emit(&cmd_buffer->batch,

View File

@@ -1470,6 +1470,32 @@ compute_pipeline_create(
vfe.CURBEAllocationSize = vfe_curbe_allocation; vfe.CURBEAllocationSize = vfe_curbe_allocation;
} }
const struct anv_shader_bin *cs_bin =
pipeline->shaders[MESA_SHADER_COMPUTE];
struct GENX(INTERFACE_DESCRIPTOR_DATA) desc = {
.KernelStartPointer = cs_bin->kernel.offset,
.SamplerCount = get_sampler_count(cs_bin),
.BindingTableEntryCount = get_binding_table_entry_count(cs_bin),
.BarrierEnable = cs_prog_data->uses_barrier,
.SharedLocalMemorySize =
encode_slm_size(GEN_GEN, cs_prog_data->base.total_shared),
#if !GEN_IS_HASWELL
.ConstantURBEntryReadOffset = 0,
#endif
.ConstantURBEntryReadLength = cs_prog_data->push.per_thread.regs,
#if GEN_GEN >= 8 || GEN_IS_HASWELL
.CrossThreadConstantDataReadLength =
cs_prog_data->push.cross_thread.regs,
#endif
.NumberofThreadsinGPGPUThreadGroup = cs_prog_data->threads,
};
GENX(INTERFACE_DESCRIPTOR_DATA_pack)(NULL,
pipeline->interface_descriptor_data,
&desc);
*pPipeline = anv_pipeline_to_handle(pipeline); *pPipeline = anv_pipeline_to_handle(pipeline);
return VK_SUCCESS; return VK_SUCCESS;